GENOMIC COMPUTING: A POTENTIAL SOLUTION TO THE DATA MINING AND PREDICTIVE MODELLING CHALLENGE TODAY ?
Douglas Bruce Kell · 2002
Currently the majority of packages require the user to write a lot of code as part of their operation. Few packages have a user interface good enough to enable relative novices to use these techniques smoothly and easily. However genomic computing’s potential for ploughing rapidly through vast amounts of data to evolve simple, human readable solutions, would appear to make it the ideal approach. The reason for its lack of wider recognition probably stems from the fact that the bulk of the real expertise still resides in the form of ‘hair shirt’ programs within university departments. Until the mainstream software vendors become fluent in this branch of computer science, the market is still open to any resourceful software development group. The Welsh software company, Aber Genomic Computing, has now made available a genomic computing package (gmax-bio) that is designed to address data mining and predictive modelling problems encountered in the life sciences and elsewhere. The reason data analysis is becoming ever more complex is that it has to deal with large numbers of data objects, each represented by tens, hundreds, or thousands of variables, as in DNA microarray analysis. The combinatorial explosion of solutions to be evaluated can thwart conventional attempts at empirical interpretation. Consider a predictive model with only 100 variables (e.g. levels of metabolite, antigen, gene expression etc). The simple problem of deciding whether or not (a simple ‘yes’ or ‘no’) to use each of these 100 variables gives 2 100 possibilities, which is about 10 30 . Considering that the lifetime of the universe is ‘only’ ~ 10 17 seconds, to find a solution for this comparatively trivial problem by random search would take more than an eternity. Fortunately nature has shown us a way – a process that is incredibly simple and yet phenomenally powerful – natural selection. Computing has an equivalent approach to solving these complex problems: Evolutionary Computing or the evolution of computer programs by methods of Darwinian selection.